WriteHuman Holds #1 as Originality.ai Shakes Up the November 2026 AI Humanizer Rankings

November's HumanizerBench results were shaped less by the humanizers than by one detector. Originality.ai's new model marked just 10 percent of the benchmark's humanized texts as human, down from 89 percent a month earlier, and almost every tool on the board lost ground because of it. WriteHuman lost some too, and still finished first of 16 tools with a composite score of 70.11, its fifth straight month at the top.

A quick note for new readers. The WriteHuman team runs HumanizerBench, and we hold our own tool to exactly the same rules as the other 15. Every prompt, output, detector verdict, and line of scoring code is public on GitHub, so you can check our work instead of taking our word for it.

Originality.ai changed the test this month

In late October, Originality.ai retired the detection model HumanizerBench had been testing with and launched two new ones, 5.0 and 5.0 Max. The benchmark now tests with 5.0. Instead of a simple verdict, 5.0 estimates how much of a document involves AI, and Originality says it built the new models for writing that has been edited, rewritten, or run through a humanizer. We covered the change in our Originality.ai review.

The other five detectors scored November's texts about the same as October's. But the detector score is the average of all six, so Originality's change alone took around 5 points off a typical tool's composite score, and 13 of the 14 tools tested in both months scored lower than they did in October.

Pangram also released a new model, though it changed the results much less. Two tools joined the benchmark for the first time, Ryne AI and Rephrasy, bringing the board to 16.

How the November board finished

Here is the top five.

WriteHuman's detector pass rate of 81.1% was second only to Phrasly's on the whole board, and WriteHuman kept its writing intact while getting there. It lost a single point, for one of its 30 rewrites drifting too far from the original. Phrasly lost seven.

Second in the field on five of six detectors

Against every detector except Winston AI, only one tool scored higher than WriteHuman.

Stealth Writer led on Copyleaks. On the other four, the tool ahead was Phrasly, and the way it got there matters. Phrasly was second-worst on the board for readability and second-worst for keeping the original meaning, and seven of its rewrites drifted far enough from what they were given to cost it a penalty point each.

The two strictest detectors are where the gap between tools shows most. Pangram rated 20 of WriteHuman's 30 texts as human, up from 15 in October. Originality's new model rated 9 as human, more than any tool except Phrasly. Apart from those two and StealthGPT, no tool had more than 3. Only four tools had any text rated human by both detectors, and WriteHuman had 8.

Eleven tools took penalties

HumanizerBench does not only check whether text gets past detectors. It also takes points off when a tool gets there the wrong way, flagging rewrites that drift too far from what the original said, that come back much longer than the text they were given, or that come back much shorter.

This cycle, 11 of the 16 tools lost points. Humbot, HIX Bypass, and Walter Writes each hit the 10-point cap for length inflation, and newcomer Rephrasy lost 11 points, 10 of them for meaning drift.

Around the rest of the board

StealthGPT finished second again and posted the best readability score on the board for the second month running. Detection is where it slipped. Its detector pass rate fell from 88% to 66%, the biggest drop of any returning tool, and it lost a point for a rewrite that came back much shorter than the original. Our StealthGPT comparison lays out the head-to-head.

Clever AI Humanizer moved up from fourth to third and took no penalties after losing four points to meaning drift in October. The strict detectors were its weak spot. Pangram rated just 1 of its texts as human, down from 8, and its detector pass rate fell from 80% to 67%. Our Clever AI Humanizer review has the details.

Phrasly jumped from twelfth to fifth, the only returning tool to score higher than in October, on the strength of detection alone. Our Phrasly comparison covers the trade-off.

Undetectable AI fell from third to sixth after losing seven penalty points, six of them for length inflation. See our Undetectable AI comparison.

Ryne AI debuted in eighth at 61.24 with no penalties. It kept the original meaning well but did not get past the strict detectors, and neither Pangram nor Originality rated any of its texts as human. Our Ryne AI review covers its plans and pricing.

Rephrasy debuted in last place at 52.50. Its detector pass rate was third-best on the board, but it kept less of the original meaning than any other tool.

The full November standings

For anyone who wants the formula, the composite score weighs detector results at 42%, meaning preservation at 32%, readability at 16%, and consistency at 10%, with penalty points subtracted after that. The full methodology, every sample, and every detector verdict are public at humanizerbench.com.

For profiles and pricing of every tool on the board, see our guide to the best AI humanizers of 2026.

What to watch in December

The question for December is whether the rest of the field adapts to Originality's new model or keeps losing points to it. WriteHuman had more texts rated human by both Pangram and Originality than any tool except Phrasly, which is a strong place to start from. The December results will be posted here as soon as they are in.

Rankings only tell you so much about your own writing. Draft something with AI, run it through WriteHuman, check the result with our AI detector, and see whether it still reads like you wrote it.